A quick and free library for MT4, much to the delight of neuralnetworkers - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Z.I. Although my EA hasn't had such drastically different results, it could depend on many conditions.
Does your EA use a different indicator for input?
The initial weights may be different, but the end result on the same input data should be about the same training results.
Otherwise it's hard randomness, not training.
The 2nd optimisation run was done from "0" on the same data with the same parameters.
There is nothing to "learn", as it is a new training, but it gives completely different results.
With new grid training, the initial weights are initialised with random numbers. Since the initial weights are different, the training for the meshes will also be different and the final result will also be different.
What is the sense in deleting ready trained grids and retraining them again?
I only optimize the training after losing trades. In most cases, StopLoss remains the same and the Expert Advisor only needs to reload the terminal to load fresh grids, because the loading from files is located in init(). If StopLoss is changed, it should be changed in the settings of the trading advisor.
Re-optimisation can be done on the fly, i.e. if the EAs are trading at the same time, the optimiser does not interfere with them.
Does your EA use a different indicator for entry?
The inputs are different, but much more important is probably the fact that the learning period is much longer...
Hello Yuri.
If you would be so kind as to explain why there are two separate optimisations for your EA,
...x: 1 to 5000 step 1
...
The initial weights may be different, but the end result on the same input data should be about the same training results.
Otherwise it's hard randomness, not training.
Don't find it difficult to read the EA's instructions carefully about what the values for x should be when optimising and don't ask any more silly questions.
If you're doing your own thing, then:
Don't blame the mirror when you're wrong (from Kozma Prutkov)
When a new grid is trained, the initial weights are initialized with random numbers.
Since the initial weights are different, the training for the grids will also be different and the final result will also be different.
What is the point of deleting ready trained grids and retraining them again?I only perform optimization - grid training - after losing trades. Most often StopLoss remains the same and for the EA to load fresh grids, you only need to reload the terminal, because the loading from the files is in init(). If StopLoss is changed, you need to change it in the settings of the trading EA as well.
Re-optimization can be done on the fly, i.e. if the Expert Advisor is trading at the time, the optimizer does not interfere with it.
The example shows initial parameters of 2 different strategies, trained on the same data, same period, with the same training parameters, for the same Expert Advisor.
From this example, it follows that if initial parameters of 2 identically trained networks are different, then their subsequent training (improvement) will be different.
If I understood correctly, then all next trainings (optimizations) should be performed without removing data from ANN directory, just extending the optimization interval by adding a new period (one day or week), i.e. the start date of the interval should not change during a new optimization?
Please read the EA's instructions carefully to see what the values for x should be during optimisation and don't ask any more stupid questions.
If you're doing it for yourself, then:
Don't blame the mirror when you're wrong (from Kozma Prutkov)
I'm sorry, I didn't quite understand that x from 0 to 1000000 is a tough condition.
I'm sorry, I didn't realise that x from 0 to 1000000 is a hard condition.
You don't have to ask or ask anything.
Normal people do everything according to the instructions first, and then experiment, if competence is enough, at their own risk.
The abnormal ones do it all at once through the... and then they start to complain right away.
mgribachev писал(а) >>
If I understood correctly, then all the subsequent trainings (optimisations) should be carried out without deleting data from the ANN catalogue, just extending the optimization interval by adding a new period (one day or week), i.e. the start date of the interval should remain unchanged during the new optimization?
I don't even know how to answer to make it more relevant? Try answering your question first:
If you want to watch another programme on TV, what would the actions be:
1. Watch on the previous television?
2. Remove the old TV, buy a new one and watch on it?
Read the EA manual again carefully to see if it mentions removing the grids.
Can someone explain something in the examples from FANN? In particular, in the file cascade_train.c the following stuff is written:
I probably don't understand something, but it seems that each subsequent pair of training and test file uploads will overwrite the previous ones. Why is this happening? Moreover, there is a memory leak because fann_destroy_train(data) is not called between loads;Then there's more:
What's the point?Someone uncommented all the options. There are no such errors in the code from CVS.